# Benchmark Radar

> Benchmark Radar is a daily-updated tracker and searchable library of public AI benchmarks. It covers emerging benchmark releases, established evaluations, public attention signals, model-lab adoption evidence, and monthly benchmark activity by field.

## Primary pages

- [Benchmark Radar](https://benchmark-radar.com/): Recent public benchmark releases ranked by observable attention.
- [Benchmark Library](https://benchmark-radar.com/#library): Searchable catalogue of public and established AI benchmarks.
- [Benchmark Trends](https://benchmark-radar.com/#trends): Monthly release activity and benchmarks shaping each field.
- [Methodology](https://github.com/Claire1217/benchmark-radar/blob/main/docs/METHODOLOGY.md): Admission, ranking, and data-quality rules.
- [Complete catalogue](https://github.com/Claire1217/benchmark-radar/blob/main/AWESOME_BENCHMARKS.md): Source-linked benchmark index.

## Machine-readable data

- [Recent benchmark index](https://benchmark-radar.com/data/benchmarks_index.json): Recent releases, resource links, descriptions, attention observations, and time-window rankings.
- [Benchmark library index](https://benchmark-radar.com/data/library_index.json): Established benchmark records, taxonomy, availability, and tracked model-report use.
- [Domain trends](https://benchmark-radar.com/data/domain_trends.json): Monthly benchmark-family counts and field-level tracked-use summaries.

## Interpretation

- Radar records represent public, reusable evaluation benchmarks rather than papers that only report findings on existing benchmarks.
- Public attention uses observable Hugging Face votes, dedicated-repository GitHub stars, and dataset downloads. Missing signals remain unknown rather than becoming zero.
- Library ranking emphasizes tracked model-lab use and public evaluation coverage.
- Monthly benchmark release volume indicates evaluation activity, not proof that a field is close to deployment.
- Benchmark Radar is a discovery index, not a benchmark-quality endorsement or a model leaderboard.

## Attribution and reuse

When citing a benchmark, follow its linked paper, project, code, or dataset and cite the original source. Public access to Benchmark Radar does not by itself grant reuse rights to repository code, compiled metadata, or third-party benchmark materials.